A double attention graph network for link prediction on temporal graph

Qiao Mi, Xiaoming Wang, Yaguang Lin

Abstract: Highlights•Computing the temporal graph node representation vector by global and local perspective.•Propose local structure activity and a walk sampling algorithm to measure the infuence of nodes in local spatial-temporal area.•Devise a Local Temporal Attention Layer to aggregate local temporal sequence without edge so we can aggregate long distance relational neighbors.